VIDEO PRE-PROCESSING OF IMAGE
INFORMATION FOR VEHICLE
IDENTIFICATION
Dr.G.Padmavathi
Department of Computer Science, Avinashilingam Deemed University for Women, Coimbatore, Tamil Nadu 641101, India
Mrs.D.Shanmugapriya
Department of Information Technology, Avinashilingam Deemed University for Women, Coimbatore, Tamil Nadu 641101, India
Ms.M.Kalaivani
Department of Computer Science, Avinashilingam Deemed University for Women, Coimbatore, Tamil Nadu 641101, India
Abstract:
This paper presents a video-based vehicle identification system which consists of preprocessing, segmentation, object extraction, object tracking and vehicle classification. The linear and non-linear filtering techniques are adopted here to filtering the video sequences for noise removal. Image filtering algorithms are applied on videos to remove the different types of noise that are either present in the video during capturing or injected in to the video during transmission. In this work the different filtering algorithms are adopted and the performances of the filters are compared using Peak Signal to Noise Ratio (PSNR) and Mean Square Error (MSE).
Keywords: vehicle Identification, video processing, Linear Filters, Non Linear Filters, Performance Metrics.
1. Introduction
Preprocessing consists of those operations that prepare data for subsequent analysis that attempts to correct or compensate for systematic errors. The digital imageries are subjected to several corrections such as geometric, radiometric and atmospheric, though all these corrections might not be necessarily are applied in all cases. These errors are systematic and can be removed before they reach the user. The investigator should decide which pre-processing techniques are relevant on the basis of the nature of the information to be extracted from remotely sensed data. Clearly, a poor contrast is expected between targets and surrounding, as the temperature differences are very low. Thereby after image acquisition, some pre-processing is required to enhance specific image features. Image enhancement is performed to improve the raw image by suppressing noise and to emphasis structures [3][4]. This will make segmentation easier.
As an image enhancement technique often drastically alters the original numeric data, it is normally used only for visual (manual) interpretation and not for further numeric analysis. Common enhancements include
• Image reduction, • Image rectification, • Image magnification, • Transect extraction, • Contrast adjustments, • Band rationing, • Spatial filtering, • Fourier transformations, • Principal component analysis and • Texture transformation.
to a degree of error, and noise is added to each image. In the same way, in the acoustic equipments used for image acquisition some noise is present internally or during transmission the noise can be added to an image. To reduce the noise significant work has been done in both hardware and software for improving the signal-to-noise ratio in acoustic images.
1.1.Significance of pre-processing
The pre-processing is required for any kind of data to refine for further processing due to poor captured quality. The following reasons justify why the pre-processing is necessary.
i. Image degradation is due to specific transmission properties of light like absorption and scattering.
ii. Specificity of environment like light changing, Weather condition, and hue is more or less predominant when vehicles move.
iii. Specificity of video captures like unknown rigid scene and unknown color or low light sensitivity.
Therefore an attempt has been made to identify the suitable filter for pre processing the video sequences frames. This work is organized as follows. Section.2 describes the filtering techniques that are widely available for video pre processing [2], Section.3 represents the experiments and results. Finally Section 4 gives the conclusion. The quantitative results of comparison are also tabulated by calculating the PSNR and MSE of the output vehicle video frame. It also provides a future scope.
2. Filtering Techniques
Filters decrease the noise by diminishing statistical deviations. It helps to reduce image noise. When an image is captured or acquired it is affected by some noise. So when image is processed, filtering techniques are applied. The main objective of filter is to reduce noise in image and to enhance the input image. It is a necessary step to improve image quality before further analysis and processing, e.g. image segmentation and restoration, can be achieved. The purpose of filtering is to reduce noise level while preserving principal signal features, such as edge contours and line details [1][6]. There are two different type of filtering techniques in basic, they are,
i. Linear Filters ii. Non Linear Filters
2.1.Linear Filter
Low pass filtering involves the elimination of the high frequency components in the image. Linear filtering can improve images in many ways: sharpening the edges of objects, reducing random noise, correcting for unequal illumination, and not convolution to correct for blur and motion, etc. These procedures are carried out by convolving the original image with an appropriate filter kernel, producing the filtered image. A serious problem with image convolution is the enormous number of calculations that need to be performed, often resulting in unacceptably long execution times. Low pass filter lack the capabilities for performing image analysis.
2.2. Non Linear Filter
The non-linear filters applied in images are required to preserve edges and details, and remove Gaussian and impulsive noise.
In this paper, a study is made on the various denoising algorithms and the algorithms are implemented in Matlab 7.9.0. Each method is compared and classified in terms of the metrics like Peak Signal to Noise Ratio (PSNR) and Mean Square Error (MSE). In order to quantify the performance of the various denoising algorithms, high quality images are taken and some known noise is added to them. Then the modified image is given as input to the denoising algorithm, which produces an image close to the original high quality image. The performance of each algorithm is compared by computing PSNR and MSE ratios.
Figure 1: Filter classification
Noise removal or noise reduction can be done on video or image using any one of these techniques. Each technique has its own advantages and disadvantages.
3. Results and Discussion
The performance of the seven filters of linear type and five filters of non linear type for vehicle video sequences are evaluated using two metrics namely PSNR and MSE. The video sequence taken for study is given in Annexure-I as frames. The PSNR value must be high and MSE value should be low for good image. During the reconstruction of a frame, each frame is assumed to have the dimension of 256. The video frames in this contain a wide variety of subject matters and textures. The quality of a video frame is examined by objective evaluation as well as subjective evaluation. For subjective evaluation, the video frames have to be observed by a human expert. The human visual system (HVS) is so complicated that it is not yet modeled properly. Therefore, in addition to objective evaluation, the video frame must be observed by a human expert to judge its quality. The following metrics are used for evaluation. They are,
The metric MSE in defined in Eq. (1):
1 1
2
0 0
1
( , )
( , )
m n
i j
MSE
I i j
K i j
mn
--- (1)
For two m×n monochrome images I and K, one of the images is considered a noisy approximation of the other. Other metrics like RMSE, MAE and PSNR are defined using MSE.
The metric Peak Signal to Noise Ratio (PSNR) is defined in Eq. 2:
2 10
10
10.log
20.log
I
I
MAX
PSNR
MSE
MAX
MSE
--- (2)
between 30 and 50 dB. The higher the value, the better the images are. Acceptable values for wireless transmission quality loss are considered to be about 20 dB to 25 dB. PSNR and MSE value is calculated between the original image and the filtered images. The values are used to evaluate and to find out the best suited filter which gives the better performance for vehicle video sequence. Here, wide variety of filtering algorithms have been implemented to detect and remove noise, leaving as much of possible of the pure signal. As the vehicle video frames exhibit more noise than visible and light ones spatial filters is robustly used. On the experimental results the Gaussian filter is very effective against the Impulse noise. Thus there is always a need for robust noise eliminating algorithm that has the capability to analyze the pattern of distribution of the pixels to separate pixels of the vehicle from background pixels. Here the linear and non linear noise removal methods are evaluated with performance metrics like MSE and PSNR. Figure 2 shows the subjective analysis of linear filtering techniques of frame 2 of the vehicle video sequence.
(a) Original Video Frame (b) After Homomorphic Filtering
(a) Original Video Frame (b) After Anistrophic Filtering
(a) Original Video Frame (b) After Gaussian Filtering
(a) Original Video Frame (b) After Circular Average Filtering
(a) Original Video Frame (b) After Unsharp Masking Filtering
(a) Original Video Frame (b) After Wavelet by Average Filtering
Comparison of PSNR Values for various Linear Filters 20.4523 19.9436 6.3571 41.8086 32.1924 24.5452 18.4782 0 5 10 15 20 25 30 35 40 45 Moro pho logic al Anist rophi c LOG gaus sian circu
lar av
erag
e
Unsh
arp M
ask
Wav
elet
by A
verag e Linear Filters R a ng e of V a lu e s PSNR Values
Figure 3: Comparison of PSNR values for various linear filtering techniques
Comparison of MSE values for various Linear Filters
534.3352 663.9242 1.52E+04
4.3213 39.5588 230.1227 40.3214
0 2000 4000 6000 8000 10000 12000 14000 16000 Mor opho logi cal Anist roph ic LOG gau ssian circu lar av
erag e UM F Wa vele t by Ave rag e Linear Filters R a nge o f V a lu e s MSE Values
Figure 4: Comparison of MSE values for various linear filtering techniques
From the above analysis it is clear that the Gaussian filter performs well in vehicle video preprocessing. The non linear filtering techniques are discussed below. Figure 5 shows the subjective analysis of the non linear filtering techniques.
(a) Original Video Frame (b) After Component Median Filtering
(a) Original Video Frame (b) After Vector Median Filtering
(a) Original Video Frame (b) After Modified Spatial Median Filtering
Figure 5: subjective analysis of non linear filtering techniques of frame 2 of the vehicle video sequence.
Comparison of PSNR values for various Non Linear Filtering
4.10E+02
550.2677 525.1728571.9073
473.8608 0.00E+00 1.00E+02 2.00E+02 3.00E+02 4.00E+02 5.00E+02 6.00E+02 7.00E+02
Median CMF VMF SMF MSMF
Non Linear Filters
R a nge of V a lue s PSNR Values
Figure 6: Comparison of PSNR values for various non linear filtering techniques
Comparison of MSE Values for various Non Linear Filtering
37.9942
20.7251 21.2357 20.5581 21.5906
0 5 10 15 20 25 30 35 40
Median CMF VMF SMF MSMF
Non Linear Filters
R a nge of V a lue s MSE Values
Figure 7: Comparison of MSE values for various non linear filtering techniques
4. Conclusion
The vehicle video sequence frames basically suffer from transmission properties, the transmission of limited range of light, disturbance of lightening, low contrast and blurring of frame, color diminished during capturing and etc. Nowadays pre-processing is only done for correcting the non-uniform lights or color and intensity adjustment. There is also a requirement for additional knowledge about the environment during pre-processing. The filters discussed in this paper are linear and non linear filtering techniques. Among these two filters, the performances of the filters are compared and analyzed by the PSNR and MSE for vehicle video sequence frames. The resulting filters which improve the image quality, suppress the noise, preserves the edges in a frame, enhance and smoothen the frame. The performance of filtering are calculated by the mean square error value which must be low for an image and peak signal to noise ratio which must be high in an image. The results are compared and concluded that Gaussian Filer in linear filtering and Spatial Median filtering in non linear filtering is best for this vehicle video sequences.
Acknowledgments
The authors would like to thank the Armament Research Board (ARMREB-DRDO) for supporting this Research project by funding.
References
[1] C. Andrew SegaU and Aggelos K. Katsaggelos, "Pre- and Post-Processing Algorithms for Compressed Video Enhancement," IEEE Explorer, 2000
[2] C.A. Segall and A.K. Katsaggelos, “Enhancement of Compressed Video using Visual Quality Metria,” Proceedings of the IEEE
Intemational Conference on Image Processing, Vancouver, BC, Canada, Sept. 10-13, 2000.
[3] C.A. Segall, P. Karunaratne and A.K. Katsaggelos, “Pre-Processing of Compressed Video,” Proceedings of the SPIE Conference on Visual Communications and Image Processing, San Jose, CA, Jan. 21-26,2001.
[4] James C. Church, Yixin Chen, and Stephen V. Rice Department of Computer and Information Science, University of Mississippi, “A Spatial Median Filter for Noise Removal in Digital Images”, 978-1-4244-1884-8/08/$25.00 ©2008 IEEE PP-618-623
[5] Mao-quan Li and Zheng-quan Xu, "An adaptive preprocessing algorithm for low bitrate video coding," Journal of Zhejiang University - Science A Volume 7, Number 12, 2057-2062, 2006
[6] Olgierd Stankiewicz ,Antoni Roszak, "Temporal noise reduction for preprocessing of video streams in monitoring systems", IEEE Explorer 2000
[7] P. R. Schrater, D. C. Knill and E. P. Simoncelli” Mechanisms of visual motion Detection” Nature America 2000
frame 2 frame 8 frame 15
frame 22 frame 27 frame 31
